Where the money leaks
The till knows about the sale. Nobody knows about what left without one
Goods leaving with the customer is a physical event. The receipt is an event in a system. While nobody matches the two, the gap between them is the loss — and its size is unknown to the company.
Given without a receipt
A cup, a portion, an item went to the customer and the sale was never rung up. One case is invisible; a habit only shows at month end, in the stock count.
Recordings exist, nobody reviews them
Cameras record around the clock. Finding a single episode costs a person hours — and they are still looking blind.
The gap surfaces too late
A shortfall appears weeks later. Who, when, and on which shift can no longer be reconstructed.
Peaks are known by feel
How many hand-overs an hour there really are, when the queue forms, where the bottleneck is — decided by impression, not by numbers.
What’s inside
Not “a camera with AI” — a counter you can argue with
Separate counters
The system distinguishes object types and counts only what belongs to a sale. The rest it sees but excludes — and that is visible on screen.
A line and a direction
The hand-over zone is mapped to a specific till. Crossing the line one way is a hand-over, the other way a return.
Shift monitor
Live video with detection boxes and counters. The operator can see that the system counts exactly what it should.
A clip attached to the event
For any disputed event the system pulls the matching segment from the recorder by timestamp — no manual searching.
A second look at the doubtful
Uncertain frames are re-checked by a separate model on our server instead of being dumped on a person in bulk.
A dropped link loses nothing
The on-site device keeps a queue of events and delivers it once the channel is back. A shift never falls out of the report.
Perimeter
The footage stays on site
Video of the till area and of staff faces is sensitive data. The topology is built so the stream goes nowhere and only the fact of an event leaves the site.
Recognition happens locally
The neural network runs on a device next to the camera. Frames are not sent to external services — neither ours nor anyone else’s.
A closed perimeter
Device and server are linked by an encrypted tunnel. The on-site hardware has no public ports.
Roles and an audit trail
Who may view events and clips is decided by role. Access is enforced on the server, not by hiding a button.
Your server or ours
The server side is deployed on your own hardware or a dedicated VPS. Data never leaves the chosen boundary.